ARO Index - Local AI Recommendation Research

Website

ARO Index - Local AI Recommendation Research AI Recommendation Research | tracking which businesses AI actually recommends, and publishing it.

Same message. Different sender. Opposite outcome.948 unsolicited emails sent between March 31 and July 30, 2026 produced...
10/14/2026

Same message. Different sender. Opposite outcome.

948 unsolicited emails sent between March 31 and July 30, 2026 produced zero replies. Six warm introductions in September 2026 produced four calls in one day. The research was unchanged. The sender changed.

A referral is a recommendation. A cold contact is a mention nobody asked for. When a buyer asks ChatGPT, Claude, Gemini, or Perplexity for a local business, the model occupies the referrer's position. The Index measures which businesses it names.

aroindex.com

September 2026 Charleston census, cold-query lens: 82 categories, 3 query variants, 4 models (ChatGPT, Claude, Gemini, P...
10/14/2026

September 2026 Charleston census, cold-query lens: 82 categories, 3 query variants, 4 models (ChatGPT, Claude, Gemini, Perplexity), live retrieval on all four, 984 reads, frozen cohort of 687 audited businesses.

6,494 business mentions were recorded. 991 resolved to cohort businesses. 227 of 687 cohort businesses were named at least once. 99 of 687 were named by the same model on at least 2 of 3 variants in the same category, the threshold Methodology V3 uses for a recommendation. 43 of 687 were named on all 3 variants by at least one model. 37 of 687 were named by all 4 models.

Of the 227 named businesses, 88 were named by one model only and 74 appeared on a single read.

The 151 on the second card is the same run restricted to the 43 categories Vol 3 compares against July. One dataset, two scopes.

Mention volume and selection are different measures. The ARO Index reports selection against a published denominator. Cohort, model strings and caveats: aroindex.com/research/model-shift-sep2026. Method: aroindex.com/methodology. Founder's full write-up: https://tagmakes.substack.com/p/cited-is-not-recommended-and-i-have

Greenville is being measured.The first 430 audits ran between September 23 and October 1 on the September 2026 model pan...
10/13/2026

Greenville is being measured.

The first 430 audits ran between September 23 and October 1 on the September 2026 model panel (ChatGPT, Claude, Gemini, Perplexity), using the same cohort rules as Charleston and Nashville: list defined first, three buyer intents, two-of-three per model.

Every live market is being re-measured on the new panel. Results publish to the leaderboard as each market completes.

aroindex.com

Recommendable is not recommended.Charleston cohort, July 2026: 688 businesses across 82 categories, defined before a sin...
10/12/2026

Recommendable is not recommended.

Charleston cohort, July 2026: 688 businesses across 82 categories, defined before a single query ran. 89 of 688 were named cold by at least one of the four models (ChatGPT, Claude, Gemini, Perplexity). About 3 of 4 surface under the direct lens, where the model is handed the business's own site.

The gap between the two lenses is the measurement. Eight slides on the cohort, the two lenses, and why the method moved with buyer behavior.

aroindex.com

For agencies serving local businesses, the measurable claim is movement, not placement.No one can guarantee that ChatGPT...
10/09/2026

For agencies serving local businesses, the measurable claim is movement, not placement.

No one can guarantee that ChatGPT, Claude, Gemini, or Perplexity will name a given business. What can be shown is where the models placed a business before a change, what changed, and where they placed it after, under a method that stayed fixed across both reads.

ARO Index publishes the method and re-measures every live market on each model panel.

aroindex.com

Why there are three ARO Index volumes in four months.Vol 1, Charleston, July 2026: 89 of 688 businesses recommended cold...
10/08/2026

Why there are three ARO Index volumes in four months.

Vol 1, Charleston, July 2026: 89 of 688 businesses recommended cold. Vol 2, Nashville: 481 businesses, same method, second market. Vol 3, September 2026 Model Shift: three of the four model strings in the panel changed between July and September (Claude unchanged), so a July versus September comparison measures a model change plus time. Vol 3 separates the two with a controlled test on both panels.

All three carry DOIs. Scores from the single-query method and the three-intent method are not comparable and are labeled as such. The correction log is public.

aroindex.com/methodology

ARO Index is research, not a tool, and the difference matters to the people who use it.A mention counter ranks what a mo...
10/07/2026

ARO Index is research, not a tool, and the difference matters to the people who use it.

A mention counter ranks what a model already said. The Index starts from a defined cohort, publishes the not-selected denominator, re-measures on every model panel, and logs its own corrections. Agencies and researchers who use the data are citing a measurement, not reading a dashboard.

Our founder's piece this week is on what that distinction cost to learn: https://tagmakes.substack.com/p/i-sent-948-cold-emails-and-got-zero

Charleston, July 2026 (ARO Index Vol 1): 688 businesses across 82 categories, cohort defined before any query ran.89 wer...
10/06/2026

Charleston, July 2026 (ARO Index Vol 1): 688 businesses across 82 categories, cohort defined before any query ran.

89 were recommended cold by at least one of ChatGPT, Claude, Gemini, or Perplexity. 12.9%. The remaining 599 (87.1%) were not named once.

Under direct audit, where the model is given the business's own site and category queries, 76.2% of the cohort surfaced. The gap between those two figures is what the Index tracks: businesses the models can recommend versus businesses the models do recommend.

Published with a DOI. Methodology: aroindex.com/methodology

Buyer behavior moved from search to ask. Instead of a keyword and a page of results, the buyer gives ChatGPT, Claude, Ge...
10/05/2026

Buyer behavior moved from search to ask. Instead of a keyword and a page of results, the buyer gives ChatGPT, Claude, Gemini, or Perplexity a full sentence and receives two or three names.

ARO Index measures which local businesses those models name when no business is planted in the question. The current method runs three buyer intents per model (direct, referral, seeking), twelve reads per business, and counts a model as recommending a business only when it names it on at least two of the three.

Methodology: aroindex.com/methodology
Founder's read on the shift: https://tagmakes.substack.com/p/you-can-lead-a-horse-to-water

Our founder started a Substack.ARO Index stays neutral by design. The leaderboard ranks what ChatGPT, Claude, Gemini, an...
10/01/2026

Our founder started a Substack.

ARO Index stays neutral by design. The leaderboard ranks what ChatGPT, Claude, Gemini, and Perplexity actually select, the methodology is published, and the corrections are public. Therese Grittner's Substack is the other side of that work: her read on the findings, with the numbers attached.

First piece: why there are already three published volumes (Charleston Vol 1, Nashville Vol 2, the September 2026 Model Shift report as Vol 3), what changed in buyer behavior and model behavior to force each one, and why a defined cohort with a published not-selected denominator is the difference between research and a list of mentions.

Link in the first comment.

Address

Charleston, SC
29455

Website

Alerts

Be the first to know and let us send you an email when ARO Index - Local AI Recommendation Research posts news and promotions. Your email address will not be used for any other purpose, and you can unsubscribe at any time.

Shortcuts

Share

Category